Overfitting

Overfitting is the phenomenon of the pedantic nerd among AI models – a system that learns so thoroughly by heart that it can no longer see the forest for the trees. Imagine a student who has memorized every exam question from the last five years down to the smallest detail, but completely fails when faced with a new, slightly modified question. That's exactly what happens with overfitting: the model learns the training data so faithfully that it even stores random fluctuations and measurement errors as 'truths'. An overfitted image recognition model might learn to recognize cats only when they're sitting on a green sofa – because that happened to be the case in the training data. The fatal consequence: while the model seemingly achieves perfect results on the training data, it fails miserably on new, unknown data. Overfitting is the curse of modern AI development and is fought with techniques like regularization, dropout, or early stopping.